AI ALIGNMENT FORUMTags
AF

Iterated Amplification

EditHistory
Discussion(0)
Help improve this page(1 flags)

Iterated Amplification is an approach to AI alignment, spearheaded by Paul Christiano. In this setup, we build powerful, aligned ML systems through a process of initially building weak aligned AIs, and recursively using each new AI to build a slightly smarter and still aligned AI. 

See also: Factored cognition. 

Posts tagged Iterated Amplification
Most Relevant
3
32Paul's research agenda FAQ
Alex Zhu
3y
31
3
35Challenges to Christiano’s capability amplification proposal
Eliezer Yudkowsky
3y
2
3
13Iterated Distillation and Amplification
Ajeya Cotra
2y
6
3
26A guide to Iterated Amplification & Debate
Rafael Harth
2mo
0
2
10AlphaGo Zero and capability amplification
Paul Christiano
2y
16
1
38My Understanding of Paul Christiano's Iterated Amplification AI Safety Research Agenda
Chi Nguyen
4mo
9
2
48Debate update: Obfuscated arguments problem
Beth Barnes
1mo
14
1
60An overview of 11 proposals for building safe advanced AI
Evan Hubinger
8mo
24
1
47Writeup: Progress on AI Safety via Debate
Beth Barnes, Paul Christiano
10mo
15
1
21Prize for probable problems
Paul Christiano
3y
0
1
23Relaxed adversarial training for inner alignment
Evan Hubinger
1y
8
1
15A comment on the IDA-AlphaGoZero metaphor; capabilities versus alignment
Alex Mennen
3y
0
1
12Corrigibility
Paul Christiano
2y
2
1
11Preface to the sequence on iterated amplification
Paul Christiano
2y
0
1
15Factored Cognition
Andreas Stuhlmüller
2y
1
Load More (15/40)
Add Posts